Not an easy LoRa to work with, but it can work.<code>v2</code> is a slight improvement, but I still think I can do better. I think the training data needs higher quality.

Nina Warink

PICS: EPOCHS: 10REG PICS: noneBLIP: 5 - 75steps: 10% 95/950 [02:16&lt;20:25, 1.43s/it, loss=0.149]steps: 20% 190/950 [04:40&lt;18:40, 1.47s/it, loss=0.141]steps: 30% 285/950 [07:04&lt;16:30, 1.49s/it, loss=0.148]steps: 40% 380/950 [09:28&lt;14:12, 1.50s/it, loss=0.141]steps: 50% 475/950 [11:52&lt;11:52, 1.50s/it, loss=0.138]steps: 60% 570/950 [14:15&lt;09:30, 1.50s/it, loss=0.141]steps: 70% 665/950 [16:39&lt;07:08, 1.50s/it, loss=0.134]steps: 80% 760/950 [19:03&lt;04:45, 1.50s/it, loss=0.135]steps: 90% 855/950 [21:27&lt;02:23, 1.51s/it, loss=0.13]steps: 100% 950/950 [24:02&lt;00:00, 1.52s/it, loss=0.132]

Files with SHA256: e9e858caeae54d17e80811bdd83149aa66c2b0b4d377ae1885e944896f5b463f